←
XtalPi521 posts
XtalPi avatar

XtalPi

@XtalPi

Engineering the way we discover. AI, robotics and a lot of chemistry.

◷ Joined April 2016▧ 237 Photos and videos
XtalPi avatar
XtalPi@XtalPi·21h

An experiment ends. The learning should carry forward. INFevo’s Recursive Scientific Discovery (RSD) framework is designed to retain evidence, methods, and results—including negative findings—so each research cycle builds on the last. The Popper Project and Karl Workbench bring that thinking into scientists’ daily work: generate and challenge hypotheses, examine supporting evidence, identify gaps, and propose experiments for review. Researchers can carry saved hypotheses and findings into later tasks as their ideas evolve. INFevo aims to bring these capabilities to the broader scientific community, helping teams turn what they learn today into sharper questions tomorrow. Follow @INFevoai for the models, experiments, and discoveries ahead. 🧬

XtalPi avatar
XtalPi@XtalPi·21h

From designing a molecule to understanding how a cell responds. INFevo’s OCOO-T models cellular responses to compounds, genetic perturbations, and cytokines within one unified architecture. IOCOO-T achieved state-of-the-art (SOTA) results against the methods compared on three public perturbation benchmarks: Tahoe100M, Replogle, and PBMC. The preprint: arxiv.org/abs/2606.12838 Pairing these capabilities with XtalPi’s AI molecular design and automated synthesis could bridge molecular design and cellular biology—helping researchers investigate interactions across targets and pathways, explore multi-target therapeutic strategies, and make better-informed drug discovery decisions.

XtalPi avatar
XtalPi@XtalPi·21h

Meet @INFevoai, our incubated spin-off—and congratulations on closing its Pre-A! $15M+. Two funding rounds in five months. A virtual cell model already posting state-of-the-art benchmark results. Built around drug discovery needs from day one, INFevo combines virtual cell models + automated high-throughput experiments to help researchers investigate potential targets, understand cellular responses, and prioritize what to test next. The goal: make every experiment inform the next discovery. Follow @INFevoai as the team builds and puts that ambition to the test. 👇

Post attachment

What happens inside a cell when you perturb a gene or add a drug - & how do we test those predictions at scale? INFevo is building AI virtual cell models connected to automated wet labs. We’ve closed a $15M+ Pre-A, incubated by @XtalPi. Read more: linkedin.com/pulse/infevo-ra…

XtalPi avatar
XtalPi@XtalPi·Nov 16

Exciting work from XtalPi-incubated Foundry Bio: ProTEUS, a world model for closed-loop protein evolution that treats protein variants as states and mutations as actions. The team reports leading performance across zero-shot, fine-tuning and active-learning tasks. Learning from experimental feedback, ProTEUS guided levansucrase engineering to 14x wild-type activity with fewer than 100 datapoints.

Introducing ProTEUS, our world model for closed-loop protein evolution. ProTEUS predicts how mutations affect protein function and learns from experimental feedback to guide protein design. From prediction to function. Paper & code in the replies. 🧬 #Foundrybio #science

2:23
XtalPi avatar
XtalPi@XtalPi·Nov 16

INFevo’s Recursive Scientific Discovery framework is designed to reuse hypotheses, methods and results to improve future hypotheses and experiments. Karl Workbench, running on #DeepSeek, achieved SOTA in single-cell and spatial omics in an October SciAgentArena re-evaluation.

Post attachment
XtalPi avatar
XtalPi@XtalPi·Nov 16

INFevo’s cellular models complement XtalPi’s molecular design and automated synthesis. Cellular predictions inform which targets and molecules to investigate, while experiments reveal how those molecules affect cells and provide evidence to guide the next stage of research.

XtalPi avatar
XtalPi@XtalPi·Nov 13

For those who haven’t seen it - Signet Therapeutics, which XtalPi incubated and partnered with on SIGX1094, was nominated for the Prix Galien USA Best Biotechnology Award for the program. Full write-up: en.xtalpi.com/signet-therape…

XtalPi avatar
XtalPi@XtalPi·Nov 12

The images show one readout: organoids shifted from grape-like clusters toward a more rounded shape with FAK/SRC inhibition. Alongside growth and signaling measurements, results from these experiments informed XtalPi’s next round of molecular designs.

XtalPi avatar
XtalPi@XtalPi·Nov 12

The organoid work supported targeting FAK and SRC together: blocking FAK alone could leave tumor-driving signals active. XtalPi’s AI molecular design focused on compounds that could inhibit both, with their effects tested in Signet’s gastric cancer organoids.

XtalPi avatar
XtalPi@XtalPi·Nov 12

SIGX1094 is a clinical-stage, first-in-class drug candidate for diffuse gastric cancer, an aggressive disease that's hard to model in the lab. To study its biology, Signet and XtalPi used organoids that captured key disease features and tested AI-designed molecules against them.

XtalPi avatar
XtalPi@XtalPi·Nov 12

A better disease model matters as much as a bigger AI model. With Signet, we paired AI with cancer organoids (lab-grown 3D tumor models), reaching PCC in just 6 months after making ~200 molecules, rather than thousands. SIGX1094 is now in the clinic. What did the organoids add?

Post attachment
XtalPi avatar
XtalPi@XtalPi·Nov 11

XtalPi's VAST is a multi-billion compound library with >85% synthesis success rate. Miroslav (Deep MedChem) just put it on a zoomable map so you can actually see the space. Explore the library: aifchem.com/vast

Introducing SynthonMap: a zoomable map of billion-to-trillion-scale chemical spaces, using factorized synthons and parametric UMAP density estimation. Try at: synthonmap.mireklzicar.com/?… Demo on @XtalPi VAST 👇

0:27
XtalPi avatar
XtalPi@XtalPi·Nov 10

These conversations mean more when there’s work to do together. Professor Verdine highlighted XtalPi’s combination of AI and automated labs - capabilities we’re putting to work in our DoveTree collaboration, launched last year. Until next time!

XtalPi avatar
XtalPi@XtalPi·Nov 10

For Professor Verdine, #AI’s promise goes beyond speed: more candidates explored in parallel, cheaper tests of biological hypotheses and fewer costly failures. But realizing this takes more than adding AI to existing workflows - it means rethinking how R&D is organized.